Final Exam Study Guide - Spring 2025

This set of flashcards covers key concepts from various imaging modalities including digital radiography, CT, nuclear medicine, MRI, ultrasound, and statistical analysis for exam preparation.

Sampling in Digital x-ray imaging

ADC and nyquist shit

Typical size of detector element

50-150 micrometers

How to compute Nyquist sampling in digital X-ray imaging?

f_N = 1/2d

Typical level of radiation received by an X-ray digital detector

x-ray is like 0.1 mSv ~ 2-3 mGy

CT ~ up to 15 mSv depending on region imaged

How to compute HVL for 80 kVp?

detector at panel. Add layers of aluminum until reading is half of that without aluminum attenuation

What is Sinogram?

Raw projection data visualized over all angles for one gantry rotation.

Need for a sinogram in CT imaging

For CST. 1D Fourier Transform the sinogram to get frequency domain, which is 2D FT’ed for final image

How to quantify noise in CT imaging?

a.      Noise is calculated as the standard deviation of pixel values within an ROI. This standard deviation can be represented in HU by propagating the noise across an ROI HU == 1000 (stnd dev in - out / out) .. out is water in the phantom

b.      This calculation of noise in HU is only possible because the signal and noise both come from the same source!

What is the Hounsfield Unit (HU) scale?

a.      Unit of measure representing electron density of different materials.

HU = (μ - μ_water)/ μ_water x 1000

What is windowing in CT?

Adjusting the range of displayed Hounsfield Units to change displayed contrast… does not effect image contrast, just how we visualize it

Window width and how affect image quality?

The range of HU values displayed in CT. Affects contrast.. what appears black, white, and range in between

Window level and how affect image quality?

The midpoint of the window width, affecting the brightness of the CT image

Image artifacts in CT imaging

Beam Hardening - Wedge shaped artifact spanning the image cause by high Z object

Cone beam — undersampling in cone angle

ring — deficiency in one detector

streak — high Z

View aliasing — poor sampling

What is Helical Pitch in CT

The relative speed of the table compared to the rotation of the gantry. Higher pitch creates a more spaced out helix.

HP < 1 increases dose because rotations overlap

HP = Table speed / Detector config…. config = N*Z axis colimation = #rows x detector width

Purpose of Helical CT scan

a.      Helical exams are faster because they eliminate the start/stop time required for axial scans to capture slices.

b.      Less dose because faster and less volume exposed to field (assuming HP > 1)

Calculate CNR using ACR QC phantom?

CNR = HU_in - HU_out / stnd dev_out

Factors affecting CNR value in CT?

a.      Slice Thickness, thinner slice is better CNR because there is less scatter from a wider field, BUT MORE NOISE because smaller detector area receiving signal.. higher electronic noise. (MORE)

b.      mAs. Higher mAs means signal reaches the detector, which reduces the amount of noise in the sampled data. Higher CNR.

c.      Kernel Filter (FC), lower FC preferentially boosts low frequency signal while reducing high frequency intensities. Reducing high frequency data reduces noise, improving contrast but decreasing spatial resolution…

How to compute SNR using ACR QC phantom?

SNR = sqrt(mAs x ST)…

PHANTOM?? Just take the ROI value and divide by the provided noise value?

Factors (2) affecting SNR value of CT?

Slice Thickness, kV, mAs

Thinner slices increases contrast and Spatial Resolution because less scatter from a wider field, but increases noise because less active detector area – less photons detected means more noise. This can be mitigated by increasing dose, mAs, or the amount of signal reaching the detector. Slice Thickness and mAs together determine

SNR = sqrt(mAs x ST) THESE TWO

Evaluating spatial resolution using ACR CT approved QC Phantom

Factors affecting image spatial resolution using ACR QC phantom

Effect of changing reconstruction kernel filter on HU and Noise

Where to apply Kernel filter in CT image reconstruction? Why?

Spatial frequency domain (Fourier Space). ….. Because convolution is a simple multiplicaiton

How to compute Nyquist spatial frequency in CT images?

f_N = 1/2d

Exponential decay of radioactive source. Why exponential?

Half-life time of a radioactive source

The time required for half of the radioactive atoms in a sample to decay.

Calculating the activity of decaying radioisotopes

Why do we need a collimator in NM?

Because NM is parallel imaging, we want to only count the photons that are incident parallel to the detector. The lead alloy collimator reduces detected scattered photons, and improves image resolution and localization

Spatial resolution and collimator

Because NM is parallel imaging, we want to only count the photons that are incident parallel to the detector. The lead alloy collimator reduces detected scattered photons, and improves image resolution and localization. The size of the collimator holes will determine the spatial resolution

Why do we need energy window in NM?

Selecting photopeak energies to exclude Compton continuum, which can be over half of total counts. Energy windowing is necessary to improve image contrast to improve quantitative and functional analysis confidence.

Where to apply energy window in NM?

At the photopeak energy of the radionuclide (generally 140 keV for Tc-99m)… use a 10% margin to account for energy resolution of detector

Why do we need to do attenuation correction in NM?

To improve functional and qualitative data by increasing image resolution

What is SPECT imaging?

Single Photon Emission Computed Tomography. Uses photons produced from gamma-emitting radionuclides to create reconstructed 3D images

Need for SPECT imaging

Non invasive methodology providing functional information to determine staging of proliferative diseases. Cheaper and minimal interpolation corrections for attenuation coefificents than PET

Need for PET imaging

Non invasive methodology providing functional information to determine staging of proliferative diseases. Better image quality than SPECT, but more expensive

Attenuation correction necessity in SPECT and PET

Needed to calculate attenuation coefficients from functional image to provide better contrast, spatial information, and ultimately functional quality

Intrinsic vs. extrinsic resolution in Gamma Camera

Intrinsic resolution is the system’s resolution without a collimator… Extrinsic is with collimation

Factors affecting intrinsic and extrinsic spatial resolution in Gamma Camera

Intrinsic: Crystal thickness & properties, PMT size

Extrinsic: number of collimator holes and size

How to determine SPECT image spatial resolution using Jaszczak QC phantom?

Cold rods. ACR minimum acceptability is 7.9mm rods

How to determine SPECT image contrast using Jaszczak QC phantom?

Image contrast is determined by resolving cold spheres

Main factor affecting SPECT image resolution and contrast?

Collimator hole number and size.

Factors affecting S/N , image contrast, and image spatial resolution in MR?

SNR:

Image Contrast:

T1 and T2 weighting

Main Magnetic Field strength

Spatial Resolution:

Number of Gy encodings (determines image matrix size)

FOV

How to determine PIU, image resolution and contrast using ACR approved MRI QC phantom

PIU = 100 x [1 - (Max ROI - Min ROI / Max ROI + Min ROI)]… From that one empty looking slice

Image acquisition factors affecting PIU, Spatial res, contrast in MR ACR Phantom

Gy repetitions, RF uniformity, Gradient field uniformity, T1 & T2 weighting…

How to select image matrix in MRI?

Number of Gy repetitions determines Ky dimensions, each row in Ky builds Kx from Gx pulse

Computing Nyquist spatial frequency in MRI

Same as every other Nyquist … just k-space this time

How to compute image noise in MRI?

Take two identical images, subtract them. The result is an image of noise

Image artifact aliasing in MRI

Gibbs Artifact caused by under sampling … increase Gy encodings to improve

Geometric Distortion caused by Gradient non-uniformity

What is acoustic impedance?

A physical property of the material that measures how much resistance is encountered when sound waves travel through it…

this is what determines grayscale value

Unit of acoustic impedance

Rayl (kg/m² s)

What is difference between M-mode and Doppler ultrasound?

M-mode shows motion over time; Doppler, or Color Doppler, is a standard Brightness mode image overlaid with color indicating blood flow direction and velocity.

Factor affecting spatial resolution in ultrasound images

Wave frequency. Increasing improves spatial res

Factor affecting contrast in ultrasound images

Acoustic impedance at interface

Image artifacts in ultrasound imaging

Comet Tail - Dark tail = stone, cyst… Bright tail = Not stone (could be tumor)

Reverberation - reflection from object and transducer… also thickness of object reverberated (e.g. needle)

Side/grating lobe

Difference between Normal distribution and Poisson distribution

Normal distribution is for continuous data sets - symmetrical

Poisson is for discrete (counting events)

In practice we use Normal because The limit of Poisson IS Normal distribution

Error propagation

ONLY propagate error for random process (NOISE)… can’t propagate for CT signal, only noise.

Difference between variance and standard deviation

Variance is dispersion of individual members of a set across the mean. Stndard deviation is dispersion of the who set relative to the mean…

Variance is squared units and larger than stnd dev (b/c squared and also across mean)… While Stnd dev is same units as data set

Calculating coefficient of variation

CV = (standard deviation / mean) * 100. Can be used to compare dispersion across different data sets with different units.

What is Confidence level and how to use it?

That example from class… How many counts for 1% error at 95% confidence level